About the Role
We're building the agent platform that powers NewsBreak's next-generation AI products — from local-news synthesis agents that millions of Americans wake up to, to paid-growth agents that autonomously decide what to advertise and where. One platform, several agentic products, real users, real spend, real consequences.
We're hiring our first dedicated Agent Platform engineer to own this layer end-to-end. You'll join a small, focused team that ships weekly, works closely with product, and takes eval and observability seriously. Our codebase already runs multiple agents in production — and you'll help build what comes next.
Responsibilities
- Build the agent runtime that orchestrates context assembly, tool invocation, model routing, and workflow tracking across multiple agentic products — the foundational layer that other teams build on top of
- Design the eval and observability harness that runs thousands of agent traces per day, surfaces regressions before they ship, and turns production failures into actionable improvements
- Own the context engineering layer — retrieval, ranking, compression, and memory — that determines what makes it into a model call; contribute informed opinions on RAG vs. long-context vs. structured tool returns
- Integrate and benchmark new foundation models as they become available; make principled decisions about model selection across agents based on capability and cost
- Build user-facing surfaces — playgrounds, agent traces, control panels — and ship them to production; the internal product team relies on these tools daily
- Collaborate closely with product to take new agent concepts from early brief to working v1 in weeks
Requirements
- Demonstrable experience shipping an AI product with real users — a side project, internal tool, open-source agent, or startup MVP you can speak to concretely
- Active, hands-on familiarity with modern AI development tooling (Cursor, Claude Code, Codex, v0, or equivalents) and a clear sense of how and when to apply them
- Ability to work end-to-end independently: backend, frontend, deployment, instrumentation, and iteration — without needing a fully defined spec to get started
- Developed perspective on AI agent design: context management, tool-calling protocols, eval strategy, and the tradeoffs between fine-tuning, prompting, and scaffolding
- Strong proficiency in Python (or Go / Node) with solid backend engineering experience — APIs, databases, queues, caches — and an understanding of how system design needs evolve with scale
- Sufficient frontend capability (React or Next.js) to ship functional internal tools independently
- Product sensibility: willingness to push back when something feels off, and a habit of thinking about the end user alongside the technical architecture
Preferred Qualifications
- Experience building or operating a multi-agent system in production (orchestration, sub-agents, MCP, skills)
- Hands-on experience with eval frameworks (LM-eval, custom harnesses, LLM-as-judge), prompt iteration workflows, or fine-tuning (LoRA / RLHF / DPO)
- A public artifact — GitHub repo, technical blog, paper, or demo — that reflects how you approach problems
- Experience integrating LLMs with complex, real-world data (news, ads, geo, user behavior) at scale